Re: [ports-gcc] Unbreak x11/grantlee

On Mon, Jan 28 2019, George Koehler <kernigh@gmail.com> wrote:
> Cc Rafael Sadowksi, who replied to my earlier mail about grantlee.
>
> On Mon, 28 Jan 2019 19:16:24 +0100
> Charlene Wendling <julianaito@posteo.jp> wrote:
>
>> Hi ports!
>>
>> > http://build-failures.rhaalovely.net/sparc64/2018-12-19/x11/grantlee.log
>> > http://build-failures.rhaalovely.net/powerpc/2018-12-14/x11/grantlee.log
>>
>> With ports-gcc, it needs additional CFLAGS and LDFLAGS.
>>
>> It has been tested successfully on macppc [1].
>>
>> Charlène.
>>
>>
>> [1] http://ix.io/1zr6
>>
>>
>> Index: Makefile
>> ===================================================================
>> RCS file: /cvs/ports/x11/grantlee/Makefile,v
>> retrieving revision 1.13
>> diff -u -p -r1.13 Makefile
>> --- Makefile 24 Oct 2018 14:28:12 -0000 1.13
>> +++ Makefile 28 Jan 2019 17:50:57 -0000
>> @@ -2,7 +2,7 @@
>>
>> COMMENT = template and text document library for Qt4
>> PKG_VERSION = 0.5.1
>> -REVISION = 4
>> +REVISION = 5
>> PLIST_VERSION = ${PKG_VERSION:R}
>> DISTNAME = grantlee-${PKG_VERSION}
>> CATEGORIES = devel x11
>> @@ -36,3 +36,8 @@ CONFIGURE_ARGS = -DCMAKE_DISABLE_FIND_PA
>> SUBST_VARS = PLIST_VERSION
>>
>> .include <bsd.port.mk>
>> +
>> +.if ${CHOSEN_COMPILER} == "ports-gcc"
>> +CXXFLAGS += -pthread
>> +LDFLAGS += -lpthread
>> +.endif
>>
>
> Earlier this month, I offered a different diff for x11/grantlee:
> https://marc.info/?l=openbsd-ports&m=154715527204197&w=2
>
> The error is "undefined reference to `pthread_cancel'" when building
> a shared object (.so). My diff removed -Wl,--no-undefined from flags
> (in patch-CMakeLists_txt), so the undefined symbol isn't an error.
> The above diff instead links to libpthread, so the symbol is defined.
>
> For comparison, x11/grantlee-qt5/patch*CMake* is now linking .so files
> to both libc and libpthread.
>
> Most .so files in OpenBSD have some undefined symbols because they
> don't link to libc. The flag -Wl,--no-undefined works in some
> systems, but causes errors in OpenBSD with libc symbols like memcpy.
> I know less about pthread_cancel; the call to pthread_cancel seems to
> be in libgcc, but pthread_cancel might be a symbol in libpthread, a
> weak symbol in libestdc++, or a weak symbol in libgcc itself.

Could this error be triggered by the fact that those undefined symbols
are to be found in a .a archive which is linked into a .so?
(--no-undefined shouldn't error out with missing symbols in a .so)

> My diff (removing -Wl,--no-undefined) might not be correct. When I
> sent my diff, some dependencies of x11/grantlee were broken (on
> platforms like macppc), so there was no way to test my diff. I only
> know that my diff worked with my modified ports tree.

Thanks for sharing your analysis. For now I tend to prefer the "remove
the -Wl,--no-undefined flag" approach, because ports shouldn't have to
explicitely specify libs because of whatever weak symbols present in
whatever system-specific lib. I still feel like there's a bug lurking
in ld.bfd(1).

Charlene, what do you think? Any preference regarding which approach to
take?

--
jca | PGP : 0x1524E7EE / 5135 92C1 AD36 5293 2BDF DDCC 0DFA 74AE 1524 E7EE